CV

Jet Lin

jetlin101@gmail.com | LinkedIn | GitHub

Artifical Intelligence/Machine Learning Scientist passionate about leveraging leadership, technical educational skills, and MLOps expertise to create AI environments for scientific advancement.
Experienced in:
- (i) Efficiently synthesizing large-scale instruction-tuning datasets for post-training.
- (ii) Fine-tuning LLMs for medical domains, deployed in production-grade (currently utilized by 100+ doctors).
- (iii) Benchmarking LLMs with self-curated evaluation sets.

University of California, Merced | Merced, CA

May 2023 – Present

Large Language Models for Traditional Chinese Medicine (TCM)
- Developed data generation pipeline using LLM-based rephrasing (via Ollama), expanding the in-house dataset from 100K to 700K+ entries.
- Continue-pretrained and fine-tuned LLaMA and Qwen models with LoRA, delivering customized APIs now adopted by 100+ doctors for daily clinical use.
- Established the largest open-source benchmark for TCM-focused LLMs, built a 30K+ evaluation set, integrated it into lm-evaluation-harness, and evaluated 15+ models.
Weight and Biases (WandB) Mobile App
- Developed Swift mobile app to extend WandB for conveniently monitoring and managing LLM training experiments from a smartphone.
Tongue Syndrome Diagnosis
- Fine-tuned SigLip on crowdsourced tongue images and expert annotations to perform multi-label classification (40+ Tags) for syndrome diagnosis.
Herb Recommendation System
- Pretrained Transformer-based model to develop herbal recommendation system covering 300+ herbs, personalized by diagnosis.

October 2025 – November 2025

AgenticMath AI Tutor Development
- Conducted data audit of 12 open-source math datasets and market analysis of 5 competitors.
- Engineered proof-of-concept RAG pipeline synthesizing education-focused datasets that sends actionable feedback to the teacher.

January 2025 – May 2025

Gene Expression Analysis with Traumatic Brain Injury Prediction
- Built 5+ models (LightGBM, Catboost, LSTM RNN, TabNet, and more) to analyze Traumatic Brain Injury from gene expression data created by X10e.
- Overcame limited gene dataset by using hyperparameter tuning and feature selection to achieve a 30% accuracy performance gain from 40% baseline.
- Deployed with AWS SageMaker and integrated with Gradio UI; used by X10e to support ongoing gene expression research.

February 2024 – May 2025

Tutored students in differential equations, statistics, linear algebra, machine learning, and more, improving their conceptual understanding and academic performance.

September 2022 – May 2025

Led 180+ participant 36-hour 27 workshop hackathon, coordinated logistics, event promotion, and secured $10,000 budget from sponsors.
Created and taught 6 workshops, including one on pretraining a ChatGPT-style LLM deployed with Gradio, engaging 50+ workshop participants.
Built event management web platform in React, helped implement live tracking for 600+ event actions in collaboration with GoBadger LLC.

November 2022 – April 2024

Managed logistics, fundraising, and marketing of 9 clubs, coordinating 17+ hours of classes per week for 300+ members.

AI/ML: Fine-tuning LLMs, RAG, vLLM, Ollama, Pytorch, scikit-learn.
Programming Languages: Python, C++/C, HTML/CSS/JavaScript, MATLAB, SQL, MIPS.
Tools: Gradio, Git, React.js, AWS SageMaker, Firebase, Flask API, Vim, Tmux, Figma.